JP4451037B2 - Information search system and information search method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an information retrieval system and an information retrieval method for inferring a user's emotion from voice information spoken by the user and retrieving predetermined information based on the inferred user's emotion.
[0002]
[Prior art]
In recent years, with the development of communication networks, users of communication networks can easily acquire information that the users want to know, for example, via the Internet. In particular, an environment in which a user can always connect to a communication network by using ADSL (Asymmetric Digital Subscriber Line) or the like has recently become widespread even in ordinary households, so that users can easily obtain information at any time. be able to.
[0003]
Thereby, the user has an advantage that information matching his / her request can be easily obtained through the Internet by using a search engine built in the personal computer.
[0004]
[Problems to be solved by the invention]
However, as the amount of information that can be searched on the Internet has increased with the development of the Internet, users cannot easily obtain the information they want using the conventional search engine. It has become.
[0005]
For example, if a user wants to search for information related to a book, the user can enter a keyword of “interesting” or “book” using a conventional search engine, but a search engine Then, the search results related to “interesting” and “books” are searched for a huge amount of about 100,000 items, and the searched contents are completely related to the book information requested by the user. There was also the case that there was no garbage information called “Funny” spot of “Japan” book.
[0006]
In addition, the user can eliminate the garbage information by searching after increasing the number of keywords. However, the user must consider a large number of keywords and must input these keywords. There was an inconvenience of not being able to
[0007]
  Therefore, the present invention has been made in view of the above points, and based on the character string input by the user, the user's emotion is grasped and the senseAffectionTo provide an information retrieval system and an information retrieval method capable of retrieving information that matches a user's request by extracting a specific keyword to be expressed and performing a search based on the extracted keyword. Let it be an issue.
[0008]
[Means for Solving the Problems]
  The invention according to the present application has been made to solve the above-described problem, and includes an information management center in which information corresponding to a keyword is stored, and a terminal connected to the information management center via a communication network. An information retrieval system in which a terminal retrieves predetermined information requested by a user from an information management center, wherein the terminal is recognized by a character recognition unit that recognizes a character string input by the user and a character recognition unit. StringThe emotion level indicating the degree of emotion is determined from the phrases related to the emotion, the keyword that is the target of the emotion is extracted, the emotion level and the keywordBased on the userRelated to emotionInference search means for inferring emotion information, extracting a keyword associated with the inferred emotion information, and searching for information related to the extracted keyword;Create a question sentence to the user based on the keyword and output itIt is characterized by having.
[0009]
  According to the invention according to the present application, the character recognition unit of the terminal recognizes the character string input from the user, and the inference search unit of the terminal holds the user based on the recognized character string. EmotionRelated toInferring emotion information, creating a question sentence to the user, inferring emotion information held by the user based on a character string input from the user and responding to the question sentence, and to the inferred emotion information Since the associated keyword is extracted and information related to the extracted keyword is searched, the terminal can acquire the keyword characterized by the emotion held by the user. By using it, information that matches the user's request can be searched.The answer sentence creating means creates and outputs a question sentence to the user based on the keyword.
[0010]
  In addition, the userCharacter recognition meansOn the other hand, when a specific character string is input, for example, the user inputs a specific character string by speaking, or the user inputs a specific character string through the operation unit, so that the terminal can express the user's feelings. Because it automatically searches for keywords that are characterized and characterized by the emotions that you grasp, you do not have to think about keywords to search for specific information one by one, and you can save time and effort to enter multiple keywords Can do.
[0011]
The terminal may infer the emotion information based on the character string, an emotion level indicating the degree of emotion of the user included in the character string, or a time when the emotion level occurs. Thereby, the terminal can infer the emotion information based on the character string, the emotion level indicating the degree of emotion of the user included in the character string, or the time when the emotion level occurs. Person's feelings can be grasped more accurately.
[0012]
The emotion level may be determined based on an emphasized word included in the character string. Thereby, since the terminal can determine the emotion level based on the emphasized word included in the character string, for example, the degree of the user's emotion can be grasped by the adverb or exclamation that is the emphasized word.
[0013]
The invention according to the present application is characterized in that, when the number of keywords associated with the emotion information exceeds a predetermined number, information related to the keywords is searched. According to the invention according to the present application, since the terminal searches for information related to the keyword when the number of keywords associated with the emotion information exceeds a predetermined number, the user can search through the terminal, A search result corresponding to a predetermined keyword can be acquired sequentially.
[0014]
The terminal may search for information related to the keyword when the number of keywords associated with the emotion information having a high emotion level becomes equal to or greater than a predetermined number. As a result, the terminal searches for information related to the keyword when the number of keywords associated with emotion information with a high emotion level (for example, very interested in, etc.) becomes a predetermined number or more. The user can obtain only search results that are of interest to the user, for example, through the terminal.
[0015]
  The terminal searches for information related to the keyword when the keyword associated with the emotion information including the time when the emotion level occurs and the emotion level exceeds a predetermined number. It may be. Thereby, since the terminal searches for information related to the keyword when the keyword associated with the emotion information including the time when the emotion level occurs and the emotion level is equal to or more than a predetermined number, For example, the user can acquire only the search result of information that the user has been interested in in the past through the terminal.
  In addition, the terminal determines the emotion of the vehicle used, including not only the “degree of emotion” and “keyword” included in the character string spoken by the user, but also the “temporal system determined” by the tense interpreter. For example, it is possible to determine what kind of emotion the user has in the past, and it is also possible to grasp a change in emotion from the past to the present.
   Also, the AI inference part of the terminal decodes the meaning content of the Japanese spoken by the user based on the context dictionary, the lexicon relation dictionary, the phrase analysis dictionary, and the morphological analysis dictionary, and uses it from the decoded contents Inferring emotions that people have and creating sentences suitable for asking questions. In other words, the AI reasoning unit that deciphers the semantic content of Japanese, based on the deciphered semantic content, cohesiveness of the deciphered meaning, change of topic, language that forms the user's emotion, statistics of the conversation so far Inferring the emotions of the user and creating a sentence suitable for asking questions to the user.
Further, when the AI inference unit of the terminal determines that the meaning content of the decoded Japanese is an ambiguous expression, it also creates a question content corresponding to the ambiguous expression using the fuzzy function.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
[Basic configuration of information retrieval system]
An information search system according to the present invention will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram of an information search system according to the present embodiment.
[0017]
  As shown in the figure, the information search system includes an input unit 101 (acquisition means) that acquires user's voice information, and a character string corresponding to the voice information based on the voice information acquired by the input unit 101. Inferring emotion information held by the user based on the character recognition unit 102 (speech recognition unit) to be identified and the character string identified by the speech recognition unit 102, and extracting a keyword associated with the inferred emotion information, An inference engine 104 (inference search means) that searches for information related to the extracted keyword and an information management center 200 that stores information corresponding to the keyword are provided.A plurality of terminals 100 a to 100 e each including an input unit 101, a voice recognition unit 102, and an inference engine 104 are connected to the information management center 200 via the communication network 300.
[0018]
The terminals 100a to 100e infer user emotion information from contents uttered by the user, and search for predetermined information requested by the user based on the inferred user emotion information. Examples of the terminals 100a to 100e include personal computers and mobile phones.
[0019]
In this embodiment, the terminals 100a to 100e include an input unit 101, a speech recognition unit 102, a speech recognition dictionary storage unit 103, an inference engine 104, a communication unit 105, an emotion information database 106, and an output unit 107. have.
[0020]
The input unit 101 is an acquisition unit that acquires user's voice information. Specifically, the input unit 101 acquires the user's voice information and outputs the acquired voice information to the voice recognition unit 102 as a voice signal. The voice recognition unit 102 is a voice recognition unit that identifies a character string corresponding to the voice signal based on the voice information acquired by the input unit 101.
[0021]
Specifically, the speech recognition unit 102 to which the speech signal is input analyzes the input speech signal, and a character string corresponding to the analyzed speech signal is stored in a dictionary stored in the speech recognition dictionary storage unit 103. The specified character string is output to the inference engine 104 as a character string signal.
[0022]
The voice recognition dictionary storage unit 103 stores a dictionary corresponding to standard voice signals. The communication unit 105 transmits and receives data to and from the information management center 200. The output unit 107 outputs a command corresponding to an instruction from the inference engine 104, and examples thereof include a display and a speaker. Specifically, the output unit 107 to which the voice command signal is input from the inference engine 104 outputs a voice corresponding to the input voice command signal.
[0023]
  The voice recognition unit 102 is also a character recognition unit that specifies a character string input by a user through an operation unit (for example, a keyboard). In addition, voicerecognitionThe dictionary storage unit 103 also stores a dictionary corresponding to the character string input by the user through the operation unit. Thereby, the user can acquire predetermined search contents from the terminal 100 not only by speaking to the terminal 100 but also by inputting characters from the operation unit.
[0024]
The emotion information database 106 stores information related to emotions held by the user as emotion information, and examples thereof include a hard disk and a CD-ROM.
[0025]
Here, in the emotion information database 106, as shown in FIG. 7, the “emotion history table” indicating the history of the “degree of emotion” (emotion level) of the user and the “degree of emotion” of the user are tense. And a “temporal history table” indicating a history to which information on the information is added. As shown in the figure, the emotion information includes, for example, “degree of emotion”, “keyword”, “text tense”, and the like. Details of these contents will be described later.
[0026]
The inference engine 104 infers emotion information held by the user based on the character string specified by the speech recognition unit 102, extracts a keyword associated with the inferred emotion information, and information related to the extracted keyword This is an inference search means for searching.
[0027]
Here, the inference engine 104 infers emotion information based on a character string, an emotion level indicating the degree of emotion of a user included in the character string, or a time (temporal system) when the emotion level occurs. Details of this inference will be described later.
[0028]
In this embodiment, the inference engine 104 is a context dictionary for deciphering a context, a similarity dictionary for examining language similarity, a dictionary related to Japanese phrase analysis, Japanese morphological analysis (each part-of-speech, utilization form) , Classification, concatenation), based on these dictionaries, deciphering the meaning of Japanese spoken by the user, inferring the user's feelings from the interpreted meaning and using To create a sentence suitable for asking questions.
[0029]
In other words, the inference engine 104 that deciphers the meaning content of the Japanese language, based on the deciphered meaning content, the coherence of the deciphered meaning, the change of the topic, the language that forms the user's emotion, the statistics of the conversation so far Inferring the emotions of the user and creating a sentence suitable for asking questions to the user.
[0030]
The inference engine 104 is also composed of artificial intelligence and a neural network. The inference engine 104 learns a language (word, sentence, etc.) exchanged with a user from a neural network, and converts the learned language into the learned language. Based on this, it is possible to create a content for asking a question to the user.
[0031]
  Furthermore, when the reasoning engine 104 determines that the interpreted meaning of the Japanese language is an ambiguous expression, it can also create a question content corresponding to the ambiguous expression using the fuzzy function. Note that the execution of the above-described functions of the inference engine 104 is performed by an AI inference unit 104 described later.fIs mainly responsible.
[0032]
  Specifically, the inference engine 104 to which the character string signal is input from the speech recognition unit 102 determines whether the user's emotion and user are based on the elements constituting the character string corresponding to the input character string signal. The time of feelings,As a result of inference engine 104 inferring user's emotionsSort keywords contained in character strings. These “user's feelings”, “time when the user had that feeling (temporal)”
In this embodiment, “keyword included in the character string” means emotion information.
[0033]
Here, “user's emotion” (degree of emotion) includes, for example, interest / indifference, rejoicing / bottom. The “user's emotion” includes the degree of emotion that the user has, for example, very interested / interested / not interested / not interested at all. ing.
[0034]
In this embodiment, this “degree of emotion” is P1 (P; Positive) (for example, very interested) when the user has a strong positive emotion, and the user has a positive emotion. If it is simply P2 (for example, interested), if the user simply has a negative emotion N1 (N; Negative) (for example, not interested), the user has a negative emotion N2 is expressed as a case where N is strongly held.
[0035]
The “time when the user has the feeling (temporal system)” includes, for example, present, past, and future. Furthermore, examples of keywords included in the character string include spring, summer, autumn, winter, reading, and sports.
[0036]
The inference engine 104, which classifies information related to the user's emotion from the elements constituting the character string, infers the emotion held by the user based on the information related to the user's emotion.
[0037]
For example, when the inference engine 104 asks the user via the output unit 107 “What are you interested in?” And the user answers “Is there something?”, The inference engine 104 Determines that the user is not worried about what he is interested in, and asks the following questions to further deepen the question.
[0038]
The inference engine 104 responded to the user through the output unit 107 when the user asked “What was the sport doing?” And the user answered “I used to play soccer happily in the past”. In this case, the inference engine 104 infers that the user is interested in soccer relations and asks questions (conversations) one after another.
[0039]
The inference engine 104 that asks various questions to the user, based on the contents of the reply from the user, for each reply, the “user's emotion”, “degree of emotion”, “user has that emotion” The emotion information is classified into “time”, and the classified emotion information is accumulated in the emotion information database 106.
[0040]
Thereafter, the inference engine 104 searches for information related to the keyword when the number of keywords associated with the emotion information exceeds a predetermined number. The inference engine 104 can also search for information related to the keyword when the number of keywords associated with emotion information having a high emotion level exceeds a predetermined number.
[0041]
Specifically, for example, the inference engine 104 acquires keywords (P1, P2) having a high “degree of emotion” from emotion information stored in the emotion information database 106, and the acquired keywords are a predetermined number. In such a case, information corresponding to a plurality of keywords is searched from the information management center 200. This search can be performed by using a general search engine.
[0042]
In addition, the inference engine 104 searches for information related to the keyword when the number of keywords associated with emotion information including a high emotion level and a time when the high emotion level has occurred exceeds a predetermined number. You can also
[0043]
Specifically, the inference engine 104 has a higher “degree of emotion” from “the time when the user has the emotion” and “the degree of emotion” stored in the emotion information database 106 (P1, P2) If only keywords that have a feeling of “current” are acquired and the number of acquired keywords exceeds a predetermined number, information corresponding to the keyword is stored in the information management center 200. Search from.
[0044]
In addition, when the emotion information stored in the emotion information database 106 is biased to information in a certain tense (for example, the past), the inference engine 104 detects the emotion in a different tense (for example, the present) from the user. It is also possible to perform a search based on keywords that constitute the emotions that are heard.
[0045]
The inference engine 104 that has searched for information corresponding to a plurality of keywords outputs the searched information to the output unit 107, and the output unit 107 to which the information searched from the inference engine 104 is input outputs the searched information. .
[0046]
As a result, the inference engine 104 infers emotions held by the user through the output unit 107, and searches for predetermined information using the keywords derived from the inferred emotions. The matched information can be searched accurately. In addition, since the user can search for predetermined information through conversation through the inference engine 104, the user can search for predetermined information happily.
[0047]
In the present embodiment, the inference engine 104 includes a phrase recognition unit 104a, a tense interpretation unit 104b, a classification unit 104c, an emphasized word detection unit 104d, an emotion determination unit 104e, and an AI inference, as shown in FIG. Unit 104f, search execution unit 104g, and keyword detection unit 104h.
[0048]
The phrase recognition unit 104a analyzes the sentence and recognizes the meaning space of the words grasped from the sentence based on the analyzed sentence. Here, sentence analysis means analyzing sentence form elements such as parts of speech, inflection forms, classifications, and connection relations. The meaning space of words is grasped from context, sentence similarity, and sentence learning patterns.
[0049]
Furthermore, the phrase recognizing unit 104a recognizes between sentences by the above recognition. Specifically, the phrase recognizing unit 104a to which the character string signal is input from the speech recognition unit 102 that has recognized the semantic space of the word grasped from the sentence corresponds to the character string signal based on the input character string signal. Recognize between sentences.
[0050]
In this embodiment, since there is a certain time interval between sentences in this embodiment, the sentence is distinguished between sentences based on the time interval. For example, when the sentence corresponding to the character signal is “It is hot today ... Let's eat ice cream”, the phrase recognition unit 104a has a time interval in the sentence. Recognize that “is a sentence break,” and divide it into the sentence “It ’s hot today” and “Let's eat ice cream”.
[0051]
The phrase recognition unit 104a, which recognizes between sentences, divides the sentence into sentences, and uses the sentence divided into sentences as a sentence style signal 104b, a classification unit 104c, and an emphasized word detection unit 104d. And output to the emotion determination unit 104e.
[0052]
The classification unit 104c determines the type of user's emotion from the character string. Specifically, the classification unit 104c to which the style signal is input from the phrase recognition unit 104a extracts a phrase related to emotion from the elements included in the sentence based on the sentence corresponding to the input style signal, and extracts this phrase. Identify what kind of emotion the phrase belongs to.
[0053]
In this embodiment, this emotion classification is classified according to an “emotion classification table” shown in FIG. 3, for example. The emotion classification table, for example, “plus element P (P; Positive)” indicating that the user means a positive emotion element, and indicates that the user means a negative emotion element. It has a “minus element N (N; Negative)”.
[0054]
This plus element P includes, for example, “Like”, “Good”, “Relieved”, “Best”, “Rejoice”, “Worried”, etc. Further, the negative element N includes, for example, useless, bad, waiting, lowest, boring, not interested (indifference), and the like. The classification unit 104c classifies what emotions are included in one sentence based on the above “emotion classification table”, and outputs the classified result to the emotion determination unit 104e as a classification signal. .
[0055]
The emphasized word detection unit 104d extracts an element characterizing emotion strength from a character string. Specifically, the emphasized word detection unit 104d, to which the style signal is input from the phrase recognition unit 104a, has an emphasized word among the elements constituting the sentence based on the sentence corresponding to the input style signal. Whether or not is detected.
[0056]
  In this embodiment, this emphasis word can be detected according to, for example, an “emphasis word table” shown in FIG. In this "emphasis word table", as shown in the figure, for example, there are adverb and exclamation of suge, choo, uhyo, wow, hie, super, very, quite etc. included. The emphasized word detection unit 104d detects an emphasized word in one sentence based on the “emphasized word table” described above, and uses the detected emphasized word as an emphasized word detection signal as an emotion determination unit.104eOutput to.
[0057]
The emotion determination unit 104e determines the degree of emotion held by the user. Specifically, the emotion determination unit 104e, to which the classification signal, the emphasized word detection signal, and the style signal are input from the classification unit 104c, the emphasized word detection unit 104d, and the phrase recognition unit 104a, the input classification signal and the emphasized word detection signal. The degree of the user's emotion is determined based on the style signal.
[0058]
In this embodiment, the determination of the degree of emotion can be performed according to an “emotion level table” shown in FIG. 5, for example. This “emotion level table” includes, for example, a determination element (user's emotion) and “emotion level” as shown in FIG.
[0059]
The determination element means a phrase that influences the emotion of the user, and includes, for example, interest / indifference, rejoicing / bottomful, highest / lowest, relieved / covered, as shown in FIG. This determination element has the same meaning as “user's emotion” described above.
[0060]
For example, as described above, the “degree of emotion” is very interested (P1) and interested when the determination element is “interested / not interested” as shown in FIG. P2), no interest (N1), no interest (N2).
[0061]
The emotion determination unit 104e refers to the “emotion level table” based on the “emotion category” corresponding to the category signal and the emphasized word corresponding to the emphasized word detection signal, and the user's emotion grasped from one sentence Is determined, and the determined results (P1, P2, N1, N2) are output as emotion determination signals to the AI inference unit 104f.
[0062]
  For example, when the character string corresponding to the style signal is “I used to play soccer happily in the past,” the classification unit 104c detects the character string “I did it,” and the emphasized word detection unit 104d The character string “fun” is detected. EmotionJudgmentThe unit 104e refers to the table of FIG. 5 based on “having done” detected by the classification unit 104c and “fun” detected by the emphasized word detecting unit 104d, and determines the degree of emotion of the user as P1 It is judged that.
[0063]
The emotion determination unit 104e detects the keyword “soccer”, which is an important element constituting the character string, and uses the emotion information including the detection result and “degree of emotion” as an emotion information signal to the AI inference unit 104g. Output to.
[0064]
The AI inference unit 104g infers the emotion that the user has from the “emotion level” of the user determined by the emotion determination unit 104e and the keyword of the character string. Specifically, the AI inference unit 104g, to which the emotion information signal is input from the emotion determination unit 104e, is based on the emotion information corresponding to the input emotion information signal. If the keyword is “soccer”, it is inferred that the user is interested in soccer.
[0065]
For example, the AI inference unit 104g extracts a phrase (predetermined content indicating the content of inference) most relevant to the keyword from the emotion information database 106 based on a character string keyword, and extracts the extracted phrase. Phrases can be the reasoning content.
[0066]
Based on the above inference, the AI inference unit 104g further deepens the contents of the question about the field of interest, searches for keywords corresponding to the emotions the user has, and stores the keywords corresponding to the emotions in the emotion information database 106. accumulate.
[0067]
The tense interpreter interprets the time (tense) when the feeling of emotion held by the user occurs. Specifically, the tense interpreter 104b to which the style signal is input from the phrase recognition unit 104a, based on the character string corresponding to the input style signal, at which time (time) the event corresponding to the character string is generated. Interpret if it is true. In this embodiment, this interpretation is performed according to the “temporal table” shown in FIG.
[0068]
As shown in the figure, this “tense table” is composed of aspect, tense, and mood, and the tense corresponding to the character string can be specified based on these components.
[0069]
Here, the aspect A is based on a completion phase (A1) indicating that a certain event has already been completed and a continuation phase (A2) indicating that a certain event continues even when the user utters. It is composed. The tense T includes a non-past tense (T1) and a past tense (T2). Furthermore, the mood M is composed of a description method M1 (an assertion method (M2), an estimation method (M3)) and an execution method M6 (will and solicitation (M4), an instruction (M4)).
[0070]
For example, if the tense interpreter 104b interprets a character string “I used to have fun playing soccer”, refer to FIG. ), And it is determined from the characters “has been” that it corresponds to the completed phase A2 (determination (2); continuity in the past). The tense interpreter 104b that has made these determinations outputs the determined tense (determination (1), determination (2)) to the AI inference unit 104f as a tense signal.
[0071]
  The tense interpreter 104b and emotionsJudgmentThe AI inference unit 104f, to which the tense signal and the emotion information signal are input from the unit 104e, respectively, “determined tense” corresponding to the input tense signal and “emotion emotion” corresponding to the input emotion information signal. Based on the “degree” and “keyword”, the user's feelings are inferred.
[0072]
  For example, when the AI inference unit 104f infers the user's emotion from the character string “I used to play soccer happily in the past,” the tense interpreter 104b, the classification unit 104c, and the emphasis worddetectionAs a result of interpreting the element of the character string in the section 104d, the “determined tense” is the above-mentioned determination [1] (conclusive M2-past T2; something has been done in the past), determination [2] (complete phase A2; something was continued), “Emotional degree” was P1 (has been doing happily), and “Keyword” was soccer. It is inferred that there is a time when soccer (keyword) has been continued (determination [2]) in the past (determination [1]), and the user is interested in soccer.
[0073]
The AI reasoning unit 104f that made this inference infers that the user is interested in soccer in the past, and therefore, questions related to soccer are derived and the contents of the questions are deepened by successively asking the derived questions. For example, questions related to soccer include favorite soccer teams, favorite players, and the like.
[0074]
The AI inference unit 104f, which has made various questions to the user, accumulates the content of the response received from the user in the emotion information database 106 for each of “determined tense”, “degree of emotion”, and “keyword”. (See FIG. 7).
[0075]
As a result, the AI inference unit 104f determines the user's emotion including not only the “degree of emotion” and “keyword” included in the character string spoken by the user but also the “determined tense”. For example, it is possible to grasp what kind of emotion the user has in the past, and it is also possible to grasp a change in emotion from the past to the present.
[0076]
That is, the AI inference unit 104f, for example, infers that “the current emotion has not changed” when the “degree of emotion” is high in the past, and the user's current emotion Can be used to determine what emotions the current user has.
[0077]
  Furthermore, from the above, the AI reasoning unit 104f can collect many keywords included in the responses from the users, particularly keywords related to the user's emotions.CordThe row unit 104g can retrieve information requested by the user more accurately by retrieving predetermined information using a keyword related to the emotion of the user inferred by the AI inference unit 104f.
[0078]
In addition, if the user inputs character information to the AI inference unit 104f through the input unit 101 (for example, voice input, input from the operation unit), the AI inference unit 104f infers the user's emotion, Since different questions are asked according to the reasoning, it is possible to enjoy the atmosphere as if talking to a human being, and to search information about certain information more happily.
[0079]
The keyword detection unit 104h outputs a search command signal for detecting the keyword to the search execution unit 104g when the number of keywords associated with emotion information exceeds a predetermined number. The keyword detection unit 104h can also output a search command signal to the search execution unit 104g when the number of keywords associated with emotion information having a high emotion level becomes equal to or greater than a predetermined number.
[0080]
Specifically, the keyword detection unit 104h selects values (P1, P2, N1, N2) indicating “degree of emotion” from the “emotion history table” (see FIG. 7) accumulated in the emotion information database 106. ) And the keyword associated with the value.
[0081]
The keyword detection unit 104h that has acquired a value with a high “degree of emotion” and a keyword associated with the value, when the number of keywords corresponding to the acquired “degree of emotion” exceeds a predetermined number, A search command signal for searching for information (for example, a book) related to the keyword is output to the search execution unit 104g.
[0082]
For example, when there are 10 or more keywords corresponding to P1 or P2 having a high degree of emotion among the acquired “degree of emotion”, the keyword detection unit 104h searches for information related to those keywords. The search command signal is output to the search execution unit 104g.
[0083]
Further, the keyword detection unit 104h sends a search command signal to the search execution unit 104g when the number of keywords associated with emotion information including the time when the high emotion level occurs and the high emotion level exceeds a predetermined number. It can also be output.
[0084]
Specifically, the keyword detection unit 104h selects, from the “temporal history table” stored in the emotion information database 106, a value indicating “degree of emotion” for each tense, and a keyword associated with the value. To get.
[0085]
The keyword detection unit 104h that has acquired the value indicating the “degree of emotion” for each tense and the keyword associated with the value, the keyword corresponding to a certain tense among the “degree of emotion” acquired for each tense. When the value becomes a predetermined numerical value or more, a search command signal for searching for information related to those keywords is output to the search execution unit 104g.
[0086]
For example, the keyword detection unit 104h searches for information related to the keyword when the number of keywords corresponding to P1 or P2 in the past becomes five or more in the acquired “degree of emotion”. The command signal is output to the search execution unit 104g.
[0087]
Note that the keyword detection unit 104h uses the emotion information of each terminal 100a to 100e stored in the emotion information database 202 in the information management center 200 to detect the emotion of the user who uses each terminal 100a to 100e. It is also possible to detect a related keyword and output a search command signal for searching for information related to the detected keyword to the search execution unit 104g.
[0088]
Examples of keywords related to the emotions of the users who use the terminals 100a to 100e include keywords that are talked about in the age group ("World Cup" in the 20s, "If in the 50s"). Restructuring ”).
[0089]
The search execution unit 104g to which the search command signal is input from the keyword detection unit 104h, based on the keyword corresponding to the input search command signal, displays information related to the keyword in the communication unit 105, the communication unit 201, and the information management. The information is searched from the information database 204 through the unit 202, and the search result is output to the output unit.
[0090]
The information management center 200 manages user emotion information. In the present embodiment, the information management center 200 includes a communication unit 201, an information management unit 202, an emotion information database 203, and an information database 204.
[0091]
The emotion information database 203 accumulates emotion information of each user using each of the terminals 100a to 100e. This emotion information database 203 stores emotion information similar to the content shown in FIG. In addition, this emotion information database 203 may accumulate | store emotion information for every user's age group using each terminal 100a-100e and order with many keywords.
[0092]
The information database 204 stores information about search targets. Examples of this information include books, travel, and hobbies. The communication unit 201 transmits and receives data to and from the terminals 100a to 100e. The information management unit 202 controls the entire information management center 204.
[0093]
[Information retrieval method using information retrieval system]
The information search method by the information search system having the above configuration can be implemented by the following procedure. FIG. 8 is a flowchart showing the procedure of the information search method according to this embodiment. First, the input unit 101 performs a step of acquiring user's voice information (S101). Specifically, the input unit 101 acquires the user's voice information and outputs the acquired voice information to the voice recognition unit 102 as a voice signal.
[0094]
Next, the voice recognition unit 102 performs a step of specifying a character string corresponding to the voice information based on the voice information acquired by the input unit 101 (S102). Specifically, the speech recognition unit 102 to which the speech signal is input analyzes the input speech signal, and a character string corresponding to the analyzed speech signal is stored in a dictionary stored in the speech recognition dictionary storage unit 103. The specified character string is output to the inference engine 104 as a character string signal.
[0095]
Next, the inference engine 104 infers emotion information held by the user based on the character string specified by the speech recognition unit 102, extracts a keyword associated with the inferred emotion information, and relates to the extracted keyword. A step of searching for information to be performed is performed (S103). The process performed here is demonstrated based on FIG.
[0096]
In the inference engine 104, first, as shown in FIG. 9, the phrase recognition unit 104a analyzes the sentence and performs a step of recognizing the meaning space of the word grasped from the sentence based on the analyzed sentence (S200). . Next, the phrase recognition unit 104a performs a step of recognizing between sentences by the above recognition (S201). Specifically, the phrase recognizing unit 104a that has grasped the semantic space of the word grasped from the sentence recognizes between the sentence corresponding to the character signal based on the inputted character signal.
[0097]
In this embodiment, since there is a certain time interval between sentences in this embodiment, the sentence is distinguished between sentences based on the time interval. For example, when the sentence corresponding to the character signal is “It is hot today ... Let's eat ice cream”, the phrase recognition unit 104a has a time interval in the sentence. Recognize that “is a sentence break,” and divide it into the sentence “It ’s hot today” and “Let's eat ice cream”.
[0098]
The phrase recognition unit 104a, which recognizes between sentences, divides the sentence into sentences, and uses the sentence divided into sentences as a sentence style signal 104b, a classification unit 104c, and an emphasized word detection unit 104d. And output to the emotion determination unit 104e.
[0099]
Next, the classification unit 104c performs a step of determining the type of user's emotion from the character string (S202). Specifically, the classification unit 104c to which the style signal is input from the phrase recognition unit 104a extracts a phrase related to emotion from the elements included in the sentence based on the sentence corresponding to the input style signal, and extracts this phrase. Identify what kind of emotion the phrase belongs to.
[0100]
In this embodiment, this emotion classification is classified according to an “emotion classification table” shown in FIG. 3, for example. The emotion classification table, for example, “plus element P (P; Positive)” indicating that the user means a positive emotion element, and indicates that the user means a negative emotion element. It has a “minus element N (N; Negative)”.
[0101]
The classification unit 104c classifies what emotions are included in one sentence based on the above “emotion classification table”, and outputs the classified result to the emotion determination unit 104e as a classification signal. .
[0102]
Next, the emphasized word detection unit 104d performs a step of extracting an element characterizing the strength of emotion from the character string (S203). Specifically, the emphasized word detection unit 104d, to which the style signal is input from the phrase recognition unit 104a, has an emphasized word among the elements constituting the sentence based on the sentence corresponding to the input style signal. Whether or not is detected.
[0103]
In this embodiment, this emphasis word can be detected according to, for example, an “emphasis word table” shown in FIG. In this "emphasis word table", as shown in the figure, for example, there are adverb and exclamation of suge, choo, uhyo, wow, hie, super, very, quite etc. included.
[0104]
  The emphasized word detection unit 104d detects an emphasized word in one sentence based on the “emphasized word table” described above, and uses the detected emphasized word as an emphasized word detection signal as an emotion determination unit.104eOutput to.
[0105]
  Next, the emotion determination unit 104e performs a step of determining the degree of emotion held by the user (S204). Specifically, the classification unit 104c, the emphasized word detection unit 104d, the phraserecognitionThe emotion determination unit 104e to which the category signal, the emphasized word detection signal, or the style signal is input from the unit 104a determines the level of the user's emotion based on the input category signal, the emphasized word detection signal, or the style signal. To do.
[0106]
In this embodiment, the determination of the degree of emotion can be performed according to, for example, an “emotion level table” shown in FIG. This “emotion determination table” has, for example, determination elements (user emotions) and “emotion level” as shown in FIG.
[0107]
For example, as described above, the “degree of emotion” is very interested (P1) and interested when the determination element is “interested / not interested” as shown in FIG. P2), no interest (N1), no interest (N2).
[0108]
Next, the emotion determination unit 104e refers to the “emotion level table” based on the “emotion category” corresponding to the category signal and the emphasized word corresponding to the emphasized word detection signal, and the user can be grasped from one sentence The degree of the emotion is determined, and the determined results (P1, P2, N1, N2) are output to the AI inference unit 104f as emotion determination signals.
[0109]
  Specifically, when the character string corresponding to the character string signal is “I used to play soccer happily in the past”, the classification unit 104c detects the character string “I did it” and the emphasized word detection unit 104d detects the character string “fun”. EmotionJudgmentThe unit 104e refers to the table of FIG. 5 based on “having done” detected by the classification unit 104c and “fun” detected by the emphasized word detecting unit 104d, and determines the degree of emotion of the user as P1 It is judged that.
[0110]
The emotion determination unit 104e detects the keyword “soccer”, which is an important element constituting the character string, and uses the emotion information including the detection result and “degree of emotion” as an emotion information signal to the AI inference unit 104g. Output to.
[0111]
Next, the AI inference unit 104g performs a step of inferring the emotion held by the user from the “degree of emotion” of the user determined by the emotion determination unit 104e and the keyword of the character string (S205).
[0112]
Specifically, the AI inference unit 104g, to which the emotion information signal is input from the emotion determination unit 104e, is based on the emotion information corresponding to the input emotion information signal. If the keyword is “soccer”, it is inferred that the user is interested in soccer.
[0113]
Then, the AI inference unit 104g further deepens the contents of the question about the field of interest based on the above inference, searches for a keyword corresponding to the emotion that the user has, and searches the keyword corresponding to the detected emotion to the emotion information database. It accumulates in 106.
[0114]
Note that the AI inference unit 104f may infer emotions held by the user, including the time when the emotion level held by the user occurs (temporal system). Specifically, the tense interpreter 104b to which the style signal is input from the phrase recognition unit 104a, based on the character string corresponding to the input style signal, at which time (time) the event corresponding to the character string is generated. Interpret if it is true. In this embodiment, this interpretation is performed according to the “temporal table” shown in FIG.
[0115]
As shown in the figure, this “tense table” is composed of aspect, tense, and mood, and the tense corresponding to the character string can be specified based on these components.
[0116]
For example, if the tense interpreter 104b interprets a character string “I used to have fun playing soccer”, refer to FIG. ), And it is determined from the characters “has been” that it corresponds to the completed phase A2 (determination (2); continuity in the past). The tense interpreter 104b that has made these determinations outputs the determined tense (determination (1), determination (2)) to the AI inference unit 104f as a tense signal.
[0117]
  And the tense interpreter 104b and emotionJudgmentThe AI inference unit 104f, to which the tense signal and the emotion information signal are input from the unit 104e, respectively, “determined tense” corresponding to the input tense signal and “emotion emotion” corresponding to the input emotion information signal. Based on the “degree” and “keyword”, the user's feelings are inferred.
[0118]
  For example, when the AI inference unit 104f infers the user's emotion from the character string “I used to play soccer happily in the past,” the tense interpreter 104b, the classification unit 104c, and the emphasis worddetectionAs a result of interpreting the element of the character string in the section 104d, the “determined tense” is the above-mentioned determination [1] (conclusive M2-past T2; something has been done in the past), determination [2] (complete phase A2; something was continued), “Emotional degree” was P1 (has been doing happily), and “Keyword” was soccer. It is inferred that there is a time when soccer (keyword) has been continued (determination [2]) in the past (determination [1]), and the user is interested in soccer.
[0119]
The AI reasoning unit 104f that made this inference infers that the user is interested in soccer in the past, and therefore, questions related to soccer are derived and the contents of the questions are deepened by successively asking the derived questions. For example, questions related to soccer include favorite soccer teams, favorite players, and the like.
[0120]
The AI inference unit 104f, which has made various questions to the user, accumulates the content of the response received from the user in the emotion information database 106 for each of “determined tense”, “degree of emotion”, and “keyword”. To do.
[0121]
Next, when the keyword detection unit 104h has a predetermined number or more of keywords associated with emotion information, the keyword detection unit 104h outputs a search command signal for searching for information corresponding to the keyword to the search execution unit 104g. This is performed (S206). The keyword detection unit 104h may output a search command signal to the search execution unit 104g when the number of keywords associated with emotion information having a high emotion level becomes equal to or greater than a predetermined number.
[0122]
Specifically, the keyword detection unit 104h selects values (P1, P2, N1, N2) indicating “degree of emotion” from the “emotion history table” (see FIG. 7) accumulated in the emotion information database 106. ) And the keyword associated with the value.
[0123]
The keyword detection unit 104h that has acquired a value with a high “degree of emotion” and a keyword associated with the value, when the number of keywords corresponding to the acquired “degree of emotion” exceeds a predetermined number, A search command signal for searching for information (for example, a book) related to the keyword is output to the search execution unit 104g.
[0124]
For example, when the keyword detection unit 104h has 10 or more keywords corresponding to P1 or P2 in the acquired “degree of emotion”, the keyword detection unit 104h searches for information related to the keywords. Is output to the search execution unit 104g.
[0125]
Further, the keyword detection unit 104h sends a search command signal to the search execution unit 104g when the number of keywords associated with emotion information including the time when the high emotion level occurs and the high emotion level exceeds a predetermined number. It can also be output.
[0126]
Specifically, the keyword detection unit 104h selects, from the “temporal history table” stored in the emotion information database 106, a value indicating “degree of emotion” for each tense, and a keyword associated with the value. To get.
[0127]
That is, the keyword detection unit 104h that has acquired the value indicating the “degree of emotion” for each tense and the keyword associated with the value, determines the certain tense among the “degrees of emotion” stored for each tense. When the corresponding keyword becomes a predetermined numerical value or more, a search command signal for searching for information related to the keyword is output to the search execution unit 104g.
[0128]
For example, when the keyword detection unit 104h stores five or more keywords corresponding to P1 or P2 in the past in the stored “degree of emotion”, the keyword detection unit 104h searches for information related to those keywords. The search command signal is output to the search execution unit 104g.
[0129]
  The keyword detection unit 104h is provided in the emotion information database 20 in the information management center 200.3Based on the emotion information of the terminals 100a to 100e stored in the terminal, the keyword related to the emotion of the user who uses the terminals 100a to 100e is detected, and the information related to the detected keyword is searched. The search command signal can be output to the search execution unit 104g.
[0130]
Examples of keywords related to the emotions of the users who use the terminals 100a to 100e include keywords that are talked about in the age group ("World Cup" in the 20s, "If in the 50s"). Restructuring ”).
[0131]
Next, the search execution unit 104g searches for predetermined information (S207). Specifically, the search execution unit 104g to which the search command signal is input from the keyword detection unit 104h, based on the keyword corresponding to the input search command signal, transmits information related to the keyword to the communication unit 105 and the communication. The information database 204 is searched through the unit 201 and the information management unit 202, and the search result is output to the output unit.
[0132]
[Operation and Effect of Information Retrieval System and Information Retrieval Method]
According to the invention according to this embodiment, the inference engine 104 recognizes a character string input from the user, infers emotion information held by the user based on the recognized character string, and infers the inference Since the keyword associated with the emotion information is extracted and the information related to the extracted keyword is searched, the inference engine 104 acquires the keyword characterized by the emotion held by the user, and acquires the acquired keyword. By using keywords, it is possible to search for information that matches the user's request.
[0133]
In addition, the user inputs a specific character string to the inference engine 104 through the input unit 101 (for example, voice input, input from an operation unit), so that the inference engine 104 grasps the user's emotion. Because it automatically searches for keywords that are characterized by the emotions you grasp, you don't have to think about keywords to search for specific information one by one, and you can save time and effort to enter multiple keywords. .
[0134]
Further, since the inference engine 104 can infer emotion information based on a character string, an emotion level indicating the degree of emotion of the user included in the character string, or a time when the emotion level has occurred, Emotion can be grasped more accurately.
[0135]
Further, since the inference engine 104 can determine the emotion level based on the emphasized word included in the character string, the user can accurately grasp the degree of the user's emotion using, for example, the adverb or exclamation word that is the emphasized word. Can do.
[0136]
Further, since the keyword detection unit 104h searches for information related to the keyword through the search execution unit 104g when the number of keywords associated with the emotion information exceeds a predetermined number, the user can search through the keyword detection unit 104h. The search result corresponding to the predetermined keyword can be automatically acquired sequentially.
[0137]
Furthermore, when the keyword detection unit 104h has more than a predetermined number of keywords associated with emotion information with a high emotion level (for example, very interested in, etc.), the keyword detection unit 104h relates to the keyword through the search execution unit 104g. Since the information to be searched is searched, the user can acquire only search results that are of interest to the user, for example, through the keyword detection unit 104h.
[0138]
In addition, the keyword detection unit 104h is related to the keyword through the search execution unit 104g when the number of keywords associated with emotion information including the time when the high emotion level occurs and the high emotion level exceeds a predetermined number. Since the information to be searched is searched, the user can obtain only the search result of the information that has been interested in the past, for example, through the keyword detection unit 104h.
[0139]
Finally, if the user inputs character information to the AI inference unit 104f through the input unit 101 (for example, voice input, input from the operation means), the AI inference unit 104f infers the user's emotion, Since different questions are asked according to the reasoning, it is possible to feel like having a conversation with a human and to search for certain information more happily.
[0140]
【The invention's effect】
  As described above, according to the present invention, based on the character string input by the user, the user's emotion is grasped and the senseAffectionBy extracting a specific keyword to be expressed and performing a search based on the extracted keyword, it is possible to search for information that more matches the user's request.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a schematic configuration of an information search system according to an embodiment.
FIG. 2 is a block diagram showing an internal structure of the inference engine in the present embodiment.
FIG. 3 is a diagram showing the contents of an emotion classification table stored in a classification section in the present embodiment.
FIG. 4 is a diagram showing the contents of an emphasized word table stored in an emphasized word detection unit in the present embodiment.
FIG. 5 is a diagram showing the contents of an emotion level table stored in an emotion determination unit in the present embodiment.
FIG. 6 is a diagram showing the contents of a tense table stored in a tense interpreter in the present embodiment.
FIG. 7 is a diagram showing the contents of a tense history table accumulated in an emotion information database in the present embodiment.
FIG. 8 is a flowchart showing a procedure of an information search method according to the present embodiment.
FIG. 9 is a flowchart showing a procedure processed in the inference engine in the present embodiment.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 100a-100e ... Terminal, 101 ... Input part, 102 ... Speech recognition part, 103 ... Speech recognition dictionary memory | storage part, 104 ... Inference engine, 104a ... Phrase recognition part, 104b ... Temporal interpretation part, 104c ... Classification part, 104d ... Emphasis Word detection unit, 104e ... emotion determination unit, 104f ... AI inference unit, 104g ... search execution unit, 104h ... keyword detection unit, 105 ... communication unit, 106 ... emotion information database, 107 ... output unit, 200 ... information management center, 201 ... Communication unit 202 ... Information management unit 203 ... Emotion information database 204 ... Information database 300 ... Communication network

Claims (11)

An information management center storing information corresponding to the keyword and a terminal connected to the information management center via a communication network, and the terminal searches the information management center for predetermined information requested by the user. An information retrieval system that performs
The terminal
A character recognition means for recognizing a character string input by a user;
The emotion level indicating the level of emotion is determined from the phrase related to the emotion included in the character string recognized by the character recognition means, the keyword that is the target of the emotion is extracted, and based on the emotion level and the keyword Inference search means for inferring emotion information related to an emotion held by a user, extracting a keyword associated with the inferred emotion information, and searching for information related to the extracted keyword;
An information retrieval system comprising: a question sentence creating means for creating and outputting a question sentence to a user based on the keyword . The information search system according to claim 1,
The inference search means searches for information related to the keyword when the number of keywords associated with the emotion information exceeds a predetermined number . The information search system according to claim 1,
The inference search means searches for information related to the keyword when the number of keywords associated with the emotion information having a high emotion level exceeds a predetermined number . An information search system according to any one of claims 1 to 3,
The information retrieval system , wherein the emotion level is determined based on an emphasized word included in a character string . An information search system according to any one of claims 1 to 4,
The inference search means includes a tense interpreter for interpreting which tense the event corresponding to the character string corresponds to based on the character string corresponding to the input style signal . . The information search system according to claim 5,
The tense interpreter includes an aspect composed of a completion phase indicating that an event has already been completed, and an continuation phase indicating that an event continues even when the user utters, and a non-past An information retrieval system that interprets a tense corresponding to a character string by using a tense table composed of tense composed of tense and past tense, a description method, and an execution method . The information search system according to claim 1,
The inference search means includes a context dictionary for deciphering a context, a similarity dictionary for examining language similarity, a dictionary for Japanese phrase analysis, and a dictionary for Japanese morphological analysis, Sentences that are appropriate for the user to decipher the meaning of the Japanese spoken based on these dictionaries, infer the user's feelings from the decoded meanings, and for the question sentence creation means to ask the user questions An information retrieval system characterized by creating . The information search system according to claim 7,
When the inference search means determines that the interpreted meaning of the Japanese is an ambiguous expression, the AI inference unit uses the fuzzy function to determine the user's An information retrieval system characterized by judging emotions . An information management center storing information corresponding to the keyword and a terminal connected to the information management center via a communication network, and the terminal searches the information management center for predetermined information requested by the user. An information retrieval method for
A step of recognizing a character string input from a user by the character recognition means of the terminal, and an inference search means of the terminal calculates a degree of emotion from a phrase related to emotion included in the character string recognized by the character recognition means. The emotion level to be shown is determined, the keyword that is the target of the emotion is extracted, the emotion information related to the emotion that the user has based on the emotion level and the keyword is inferred, and the inferred emotion information is associated with the inferred emotion information Extracting the extracted keyword and searching for information related to the extracted keyword . The information search method according to claim 9,
An information search method comprising a step of searching for information related to the keyword when the number of keywords associated with the emotion information exceeds a predetermined number . The information search method according to claim 10,
The inference search means includes a step of searching for information related to the keyword when the number of keywords associated with the emotion information having a high emotion level exceeds a predetermined number . Method.

Images